A wind turbine generator set intelligent identification method and system

By integrating high-sensitivity acoustic sensors and environmental monitoring equipment on wind turbines and combining deep learning technology, intelligent identification and precise positioning of wind turbine faults is achieved, and the problems of fault recognition lag and insufficient environmental adaptability in traditional methods are solved. Significantly improves operation and maintenance efficiency and the accuracy of fault identification.

CN118757347BActive Publication Date: 2024-11-05STATE GRID GANSU ELECTRIC POWER CO JIUQUAN POWER SUPPLY CO +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411224460.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-03
Publication Date
2024-11-05
Estimated Expiration
2044-09-03

AI Technical Summary

Technical Problem

It is difficult to accurately judge the fault source and predict performance in real time in the operating state of wind turbines. Traditional methods are limited by the inefficiency of manual inspection and the delayed discovery of hidden faults, which affects the overall performance.

Method used

Integrate highly sensitive acoustic sensors and environmental monitoring equipment, combined with deep learning technology, through the precision processing of voiceprint data, the construction of dynamic adaptive wind energy models and the fault diagnosis strategy of multi-feature fusion, we can realize intelligent identification and precise positioning of faults, and through the environment Factor correction enhances the accuracy of fault identification.

Benefits of technology

It significantly improves the operation and maintenance efficiency of wind turbines, reduces energy losses, improves noise suppression and environmental adaptability, quickly learns new fault modes, generates detailed fault reports, and improves the accuracy and response speed of fault identification.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118757347B_ABST
    Figure CN118757347B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and system for intelligent identification of wind turbines, which relates to the field of data encryption technology, including installing acoustic sensors and environmental monitoring equipment to collect voiceprint data and environmental parameters in real time; performing noise reduction processing and feature extraction on voiceprint data, establishing a fault voiceprint database and constructing a dynamic adaptive wind energy model, adaptively adjusting model parameters, matching the extracted voiceprint features with the fault voiceprint database, and combining the output of the dynamic adaptive wind energy model to confirm the fault location and type, and analyze the impact of environmental changes on fault identification. The present invention improves the efficiency and reliability of wind power operation and maintenance through intelligence, realizes accurate fault diagnosis by optimizing data processing, feature extraction and model prediction, refines voiceprint processing and adaptive wind energy models to enhance recognition accuracy and environmental adaptability, and comprehensively analyzes to ensure immediate repair of faults and long-term health of equipment, thereby promoting the sustainable development of the wind power industry.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to an intelligent identification method and system for a wind power generator set. Background Art

[0002] As wind power generation technology develops rapidly, efficient and reliable fault diagnosis has become the key to optimizing operation and maintenance and ensuring energy output. Traditional methods are limited by the inefficiency of manual inspection and the delayed discovery of hidden faults, which affects the overall efficiency. Recently, the integration of the Internet of Things, big data and AI technologies has brought a turning point for this, but it still faces the challenges of the complexity of acoustic signals, environmental interference and the rapid identification of new faults.

[0003] The present invention integrates highly sensitive acoustic sensors and environmental monitoring equipment, combined with advanced data processing and deep learning technologies, to achieve intelligent identification and precise positioning of faults in wind turbines. The core technologies include precise processing of voiceprint data, construction of dynamic adaptive wind energy models, and fault diagnosis strategies based on multi-feature fusion, which can effectively overcome noise, improve environmental adaptability, and quickly learn new fault modes. More importantly, the accuracy of fault identification is enhanced through correction of environmental factors, and detailed fault reports are generated in real time, which significantly improves operation and maintenance efficiency and reduces energy loss. Compared with the existing technology, this solution has achieved significant progress in noise suppression, environmental adaptability and new fault learning. It is at the forefront of intelligent operation and maintenance technology for wind power generation and is of great significance to promoting the sustainable development of the industry. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for intelligently identifying a wind turbine generator set to solve the problem that it is difficult to accurately determine the source of a fault and predict performance in real time when the wind turbine generator set is in operation.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides an intelligent fault identification method for a wind turbine generator set, which includes: the wind turbine generator set is equipped with an acoustic sensor and an environmental monitoring device to collect voiceprint data and environmental parameters in real time; denoising and feature extraction are performed on the voiceprint data to obtain voiceprint features; a fault voiceprint database is established using the voiceprint features; a dynamic adaptive wind energy model and environmental parameters collected in real time are constructed based on deep learning, the dynamic adaptive wind energy model parameters are adaptively adjusted, and the wind energy environmental quality coefficient and the expected performance indicators of the wind turbine generator set under the current environment are calculated; the extracted voiceprint features are matched with the fault voiceprint database to identify abnormal voiceprint patterns; preliminary fault location is performed based on the abnormal voiceprint pattern and the dynamic adaptive wind energy model; the preliminary fault location is comprehensively analyzed with the expected performance indicators to confirm the fault location and type, and the impact of environmental changes on fault identification is analyzed.

[0008] As a preferred solution of the intelligent fault identification method for wind turbine generator set of the present invention, wherein:

[0009] The wind turbine generator set is equipped with acoustic sensors and environmental monitoring equipment to collect voiceprint data and environmental parameters in real time. The specific steps are as follows:

[0010] Voiceprint data and environmental parameters are collected in real time through high-sensitivity acoustic sensors with anti-interference capabilities installed at key locations of wind turbines and environmental monitoring equipment installed around wind turbines. All equipment is connected to the central data acquisition system.

[0011] As a preferred solution of the intelligent fault identification method for wind turbine generator set of the present invention, the specific steps are as follows: performing noise reduction processing and feature extraction on voiceprint data to obtain voiceprint features; and establishing a fault voiceprint database using voiceprint features:

[0012] The time domain filter is used to directly process the time series data of the voiceprint data, and then combined with the frequency domain analysis, FFT is used to convert the voiceprint data into a frequency domain signal. Spectral subtraction is used in the frequency domain to remove noise of specific frequencies. Finally, IFFT is used to convert the processed frequency domain signal back to the time domain to obtain the denoised voiceprint data. The expression is:

[0013] ;

[0014] Where X[k] is the kth frequency component in the frequency domain, x[n] is the input voiceprint data, k is the index of the frequency component, n represents the nth sample, and FFT means accepting a real number as input and returning the corresponding frequency domain representation;

[0015] The voiceprint features are extracted by calculating the RMS value and peak value in the time domain, spectrum analysis in the frequency domain, and Mel frequency cepstrum coefficients;

[0016] A multidimensional feature vector is constructed based on the extracted voiceprint feature vector, and principal component analysis (PCA) is used to reduce the feature dimension, eliminate redundant information and highlight key features. The expression is:

[0017] ;

[0018] Among them, X is the original voiceprint feature vector, P is the conversion matrix, Y is the voiceprint feature vector after dimensionality reduction, and T is the matrix transpose operator;

[0019] A support vector machine (SVM) model is trained using the voiceprint feature vector after dimensionality reduction, and an adaptive regularization parameter is introduced to improve the SVM model, allowing the model to automatically adjust the regularization strength according to the misclassification cost of different fault types. The expression is:

[0020] ;

[0021] Where W is the weight vector, b is the bias term, Ci is the adaptive regularization parameter, is a slack variable, m is the number of samples, and i is an index variable;

[0022] The improved SVM model is used to classify the real-time monitoring data into faults. According to the classification results, the corresponding voiceprint features and their fault types are stored in the fault voiceprint database.

[0023] The established fault voiceprint database is used to perform real-time analysis on the newly collected voiceprint data. By comparing the voiceprint features in the database, early fault detection and early warning are achieved. Trend analysis and fault cause backtracking are performed based on historical fault data. The expression is:

[0024] ;

[0025] Among them, F t+1 is the fault prediction value at time t+1, is the smoothing coefficient, is the failure prediction value at time t, is the fault prediction value at time t-1.

[0026] As a preferred solution of the intelligent fault identification method for wind turbine generator sets of the present invention, the dynamic adaptive wind energy model is constructed based on deep learning, and the parameters of the dynamic adaptive wind energy model are adaptively adjusted according to the environmental data collected in real time, and the wind energy environmental quality coefficient and the expected performance index of the wind turbine generator set under the current environment are calculated. The specific steps are as follows:

[0027] The dynamic adaptive wind energy model is constructed by combining deep learning and long short-term memory network LSTM to process time series data. The expression is:

[0028] ;

[0029] in, represents the expected performance index at time t, θ LM represents the parameter set of the dynamic adaptive wind energy model, V(t) is the wind speed time series, T(t) is the temperature time series, H(t) is the humidity time series, and θ(t) is the wind direction time series;

[0030] According to the learning mechanism in the model, the parameters are constantly fine-tuned according to new real-time data. The expression is:

[0031] ;

[0032] in, is the learning rate, is the gradient of the loss function with respect to the parameters of the long short-term memory network LSTM, is the actual measured power output, θ L,n is the updated model parameter, θ L,o are the model parameters before updating, represents the expected performance indicator at time t;

[0033] Based on model training and adaptive adjustment, the wind energy environment quality coefficient Q is defined as a composite indicator, and the expression is:

[0034] ;

[0035] Where V is wind speed, θ is wind direction, T is temperature, H is humidity, is a function based on wind speed and wind direction, β is the temperature-related attenuation coefficient, γ is the humidity-related attenuation coefficient, T0 is the reference temperature, η is a constant used to adjust the scale, is the base of natural logarithms;

[0036] Using the trained dynamic adaptive wind energy model, the current real-time environmental parameters are input, and the model output is the expected performance index of the wind turbine under the current environment. ,The output calculation of the model relies on the forward propagation mechanism of the long short-term memory network LSTM, which maps the input environmental features to the predicted values ​​of the expected performance through a series of nonlinear transformations and weight adjustments.

[0037] As a preferred solution of the intelligent fault identification method for wind turbine generator set of the present invention, the steps of matching the extracted voiceprint features with the fault voiceprint database to identify abnormal voiceprint patterns are as follows:

[0038] The fault mode feature set of the fault voiceprint database is set to {F1, F2, …, F N}, each feature vector F in the failure mode feature set i Represents a known fault mode, and uses cosine similarity as the matching metric to calculate the similarity between the new feature vector and each vector in the database. The expression is:

[0039] ;

[0040] Among them, S(i) represents the similarity between the new feature vector and the i-th fault mode feature vector, is the dot product of the vectors, F new represents the new eigenvector, F i is the feature vector of the i-th known fault mode in the fault voiceprint database;

[0041] Define an abnormal threshold T8. When S(i)< T8, it is considered that the newly collected voiceprint data does not match any known fault mode, and a new abnormal mode appears. An abnormal scoring function based on similarity is introduced, and the expression is:

[0042] ;

[0043] Among them, E is the preliminary abnormality score, (S(i)) represents the new eigenvector F new and all the fault mode feature vectors F in the database i The maximum value of the cosine similarity;

[0044] The initial anomaly score is normalized and the expression is:

[0045] ;

[0046] Among them, E min is the minimum value of the anomaly score in the historical data, E max is the maximum value of the anomaly score in the historical data, E nm is the normalized anomaly score.

[0047] As a preferred solution of the intelligent fault identification method for wind turbine generator set of the present invention, the output of the dynamic adaptive wind energy model is combined to perform preliminary fault location, and the specific steps are as follows:

[0048] Based on the anomaly score E nm Combined with the wind energy environmental quality coefficient Q, a comprehensive evaluation model is obtained, and the expression is:

[0049] ;

[0050] Among them, L is the comprehensive evaluation score, w1 is the abnormal score E nm , w2 is the weight factor of the complement (1-Q) of the wind energy environmental quality factor Q;

[0051] Three thresholds are defined: safety threshold L1, warning threshold L2, and danger threshold L3. Based on the comprehensive evaluation score L, the fault is located into predefined categories through threshold division. If L<L1, it is classified as normal; if L1≤L<L2, it is classified as attention; if L2≤L<L3, it is classified as warning; if L≥L3, it is classified as danger.

[0052] As a preferred solution of the intelligent fault identification method for wind turbine generator set of the present invention, the preliminary fault location and expected performance indicators are comprehensively analyzed to confirm the fault location and type, and analyze the impact of environmental changes on fault identification. The specific steps are as follows:

[0053] A fault diagnosis model based on multi-feature fusion is introduced, which combines the comprehensive evaluation score L and the normalized anomaly score E nm , wind energy environment quality factor Q and specific index F of fault soundprint characteristics n sp Combined to form a multi-dimensional feature vector, the expression is:

[0054] ;

[0055] Among them, F n sp is the specific index of the new eigenvector;

[0056] Introducing a correction factor based on wind energy environmental quality factor Q , reducing the misjudgment rate of fault identification due to environmental factors, the expression is:

[0057] ;

[0058] in, is a scalar amplification factor, β is a parameter that determines the steepness of the curve, Q0 is a specific environmental quality standard, is the base of natural logarithms;

[0059] The correction factor Applied to the comprehensive evaluation score L, the corrected comprehensive score is calculated as follows:

[0060]

[0061] Among them, L' is the corrected comprehensive evaluation score;

[0062] The difference ΔL between the comprehensive evaluation score L and the corrected comprehensive evaluation score L' was calculated, and the linear relationship between ΔL and each environmental parameter was tested using the Pearson correlation coefficient;

[0063] For each environmental parameter, draw a scatter plot of ΔL and the corresponding environmental parameter, add a trend line on the scatter plot to intuitively show the trend of ΔL changing with the environmental parameter;

[0064] Based on the scatter plot, determine the environmental parameters that have the greatest impact on ΔL, how changes in environmental parameters affect ΔL, and generate a fault report.

[0065] In the second aspect, the present invention provides an intelligent fault identification system for a wind turbine generator set, comprising a data acquisition module, a voiceprint data processing module, a fault voiceprint database management module, a fault identification and positioning module, and a fault report generation module. The Internet of Things data acquisition module is used to collect acoustic and environmental data of wind turbines in real time, and transmit them to a control center via a wireless network. The voiceprint data processing module is used to denoise and extract features from the collected voiceprint data to prepare pure feature information for fault diagnosis. The fault voiceprint database management module is used to maintain a comprehensive fault voiceprint feature library to support the learning and storage of new fault modes. The fault identification and positioning module is used to compare the processed voiceprint features with the database, combined with model prediction, to quickly identify and locate potential faults. The fault report generation module is used to integrate analysis results, automatically generate a report containing fault details, location and suggested countermeasures, to assist decision-making.

[0066] In a third aspect, an embodiment of the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the intelligent fault identification method for a wind turbine generator set as described in the first aspect of the present invention is implemented.

[0067] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the method for intelligent fault identification of a wind turbine generator set as described in the first aspect of the present invention is implemented.

[0068] The beneficial effects of the present invention are as follows: the present invention significantly improves the operation and maintenance efficiency and reliability of wind power equipment through the intelligentization of the entire chain from data collection, feature extraction, model prediction to fault diagnosis. Each step is optimized for specific goals. For example, the refined processing of voiceprint data enhances the accuracy of fault identification, and the introduction of dynamic adaptive wind energy models enhances the system's adaptability to environmental changes. Finally, through comprehensive analysis, fault identification is closely combined with environmental impacts. The generated fault report not only facilitates immediate maintenance, but also provides a scientific basis for long-term equipment health management, thereby promoting the sustainable development of the wind power industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0070] Figure 1 This is a flow chart of the intelligent fault identification method for wind turbine generator sets in Example 1.

[0071] Figure 2 This is a flow chart of the wind turbine generator set intelligent fault identification system module in Example 1. DETAILED DESCRIPTION

[0072] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0073] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for intelligent fault identification of a wind turbine generator set, comprising the following steps:

[0074] S1. Wind turbines are equipped with acoustic sensors and environmental monitoring equipment to collect soundprint data and environmental parameters in real time.

[0075] Through the high-sensitivity acoustic sensors with anti-interference capabilities installed at key parts of the wind turbine (such as the blade root, gearbox, and generator) and the environmental monitoring equipment installed around the wind turbine, voiceprint data and environmental parameters (temperature, humidity, wind speed) are collected in real time. All devices are connected to the central data acquisition system.

[0076] It should also be noted that through the all-round environmental monitoring devices installed around the wind farm and on the units, these devices can continuously monitor multiple environmental parameters, including but not limited to temperature, humidity and wind speed. These parameters have a direct impact on the operating efficiency and maintenance requirements of wind turbines. For example, excessive temperature may accelerate the aging of lubricating oil, humidity changes will affect the insulation performance of electrical components, and wind speed is directly related to power generation efficiency. The sensors used are all of industrial grade standards and can work stably under extreme weather conditions to ensure the continuity and integrity of the data.

[0077] S2. Perform noise reduction and feature extraction on the voiceprint data to obtain voiceprint features; and use the voiceprint features to establish a fault voiceprint database.

[0078] Furthermore, a time domain filter is used to directly process the time series data of the voiceprint data, and then combined with frequency domain analysis, FFT is used to convert the voiceprint data into a frequency domain signal. Spectral subtraction is used in the frequency domain to remove noise of a specific frequency. Finally, IFFT is used to convert the processed frequency domain signal back to the time domain to obtain the denoised voiceprint data. The expression is:

[0079] ;

[0080] Where X[k] is the kth frequency component in the frequency domain, x[n] is the input voiceprint data, k is the index of the frequency component, n represents the nth sample, and FFT means accepting a real number as input and returning the corresponding frequency domain representation;

[0081] The voiceprint features are extracted by calculating the RMS value and peak value in the time domain, spectrum analysis in the frequency domain, and Mel frequency cepstrum coefficients;

[0082] A multidimensional feature vector is constructed based on the extracted voiceprint feature vector, and principal component analysis (PCA) is used to reduce the feature dimension, eliminate redundant information and highlight key features. The expression is:

[0083] ;

[0084] Among them, X is the original voiceprint feature vector, P is the conversion matrix, Y is the voiceprint feature vector after dimensionality reduction, and T is the matrix transpose operator;

[0085] A support vector machine (SVM) model is trained using the voiceprint feature vector after dimensionality reduction, and an adaptive regularization parameter is introduced to improve the SVM model, allowing the model to automatically adjust the regularization strength according to the misclassification cost of different fault types. The expression is:

[0086] ;

[0087] Where W is the weight vector, b is the bias term, Ci is the adaptive regularization parameter, is a slack variable, m is the number of samples, and i is an index variable;

[0088] The improved SVM model is used to classify the faults of real-time monitoring data. According to the classification results, the corresponding voiceprint features and their fault types are stored in the fault voiceprint database.

[0089] The established fault voiceprint database is used to perform real-time analysis on the newly collected voiceprint data. By comparing the voiceprint features in the database, early fault detection and early warning are achieved. Trend analysis and fault cause backtracking are performed based on historical fault data. The expression is:

[0090] ;

[0091] Among them, F t+1 is the fault prediction value at time t+1, is the smoothing coefficient, is the failure prediction value at time t, is the fault prediction value at time t-1.

[0092] It should also be noted that in the voiceprint processing stage, time domain filtering and frequency domain spectrum subtraction are used for noise reduction. Through FFT and IFFT conversion, a pure signal X[k] is obtained. Feature extraction covers time domain, frequency domain characteristics and Mel frequency cepstrum, and a multidimensional vector X is constructed. After PCA dimension reduction, it is reduced to the key feature Y. The improved SVM model is trained, and adaptive regularization is introduced to dynamically adjust parameters to achieve fault classification. Using this model, the new voiceprint data is compared with the fault library for early warning. The database is dynamically updated to ensure the accuracy of prediction. The fault trend is analyzed based on historical data. Among them, the smoothing coefficient α adjusts the predicted value F t+1 , balance the impact of new and old data.

[0093] S3, build a dynamic adaptive wind energy model based on deep learning and real-time collected environmental data, adaptively adjust the parameters of the dynamic adaptive wind energy model, calculate the wind energy environmental quality coefficient and the expected performance indicators of the wind turbine under the current environment;

[0094] Furthermore, deep learning and long short-term memory network LSTM are combined to process time series data to build a dynamic adaptive wind energy model, which is expressed as:

[0095] ;

[0096] in, represents the expected performance index at time t, θ LM represents the parameter set of the dynamic adaptive wind energy model, V(t) is the wind speed time series, T(t) is the temperature time series, H(t) is the humidity time series, and θ(t) is the wind direction time series;

[0097] According to the learning mechanism in the model, the parameters are constantly fine-tuned according to new real-time data. The expression is:

[0098] ;

[0099] in, is the learning rate, is the gradient of the loss function with respect to the parameters of the long short-term memory network LSTM, is the actual measured power output, θ L,n is the updated model parameter, θ L,o are the model parameters before updating, is the expected performance indicator at time t;

[0100] On the basis of model training and adaptive adjustment, the wind energy environmental quality coefficient Q is defined as a composite index. The specific calculation method can be combined with environmental parameters such as wind speed, temperature and humidity. The expression is:

[0101] ;

[0102] Where V is wind speed, θ is wind direction, T is temperature, H is humidity, is a function based on wind speed and wind direction, β is the temperature-related attenuation coefficient, γ is the humidity-related attenuation coefficient, T0 is the reference temperature, η is a constant used to adjust the scale, is the base of natural logarithms;

[0103] It should be noted that the wind energy environmental quality factor Q is a composite index calculated by comprehensively considering wind speed, wind direction, temperature and humidity, and is used to evaluate the operating quality and performance of wind turbines under current environmental conditions;

[0104] Using the trained dynamic adaptive wind energy model, the current real-time environmental parameters are input, and the model output is the expected performance index of the wind turbine under the current environment. ,The output calculation of the model relies on the forward propagation mechanism of the long short-term memory network LSTM, which maps the input environmental features to the predicted values ​​of the expected performance through a series of nonlinear transformations and weight adjustments.

[0105] It should also be noted that through a series of nonlinear transformations and weight adjustments refers to the operations within the long short-term memory network LSTM. After receiving environmental parameters such as wind speed V(t), temperature T(t), humidity H(t) and wind direction θ(t), these data are first filtered through the input gate of the long short-term memory network LSTM to determine which information is updated to the cell state and which is ignored. This screening process involves nonlinear activation functions (such as the Sigmoid function) to ensure that only relevant information is strengthened. Next, the forget gate is responsible for determining which old information in the cell state is no longer important, which is also achieved through nonlinear transformation. After that, through the update of the cell state, combined with the new input information and some of the forgotten old information, the long short-term memory network LSTM can maintain a long-term memory state, reflecting the cumulative effect of all input data so far. Finally, through the output gate, the long short-term memory network LSTM determines which cell state information should be used for the output of the current time step. This step also involves nonlinear transformation to ensure that the output matches the task requirements. In this wind energy prediction model, the output is the expected performance indicator , which is a comprehensive result based on the current environment parameters and the internal memory state of the model.

[0106] Weight adjustment is the core of deep learning model training. The gradient of the loss function with respect to the model parameters (such as the weights and biases of the long short-term memory network LSTM) is calculated through the backpropagation algorithm , and then these parameters are adjusted according to the learning rate u, so that the model is gradually optimized to more accurately predict the expected performance indicators of the wind turbine generator set. The whole process is iterative. With each training iteration, the model parameters (θ L,o become θ L,n ) are gradually fine-tuned to better fit the actual performance data, and finally the purpose of efficiently predicting the wind energy environment quality coefficient Q and the expected performance indicator P ep (t) is achieved.

[0107] S4. Match the extracted voiceprint features with the fault voiceprint database, identify abnormal voiceprint patterns, and combine with the output of the dynamic adaptive wind energy model to perform preliminary fault location;

[0108] Furthermore, set the fault mode feature set of the fault voiceprint database as {F1, F2, …, F N}, and each feature vector F i in the fault mode feature set represents a known fault mode. The cosine similarity is used as the matching metric to calculate the similarity between the new feature vector and each vector in the database. The expression is:

[0109] ;

[0110] where S(i) represents the similarity between the new feature vector and the i-th fault mode feature vector, is the dot product of the vectors, and || || represents the Euclidean norm of the vector; F new represents the new feature vector, and F i is the feature vector of the i-th known fault mode in the fault voiceprint database;

[0111] Define an abnormal threshold T8. When S(i) < T8, it is considered that the newly collected voiceprint data does not match any known fault mode, and a new abnormal mode appears. Introduce a similarity-based abnormal scoring function. The expression is:

[0112] ;

[0113] where E is the preliminary abnormal score, (S(i)) represents the maximum value of the cosine similarity between the new feature vector F new and all the fault mode feature vectors F i in the database;

[0114] Perform normalization processing on the preliminary abnormal score. The expression is:

[0115] ;

[0116] Among them, E min is the minimum value of the anomaly score in the historical data, E max is the maximum value of the anomaly score in the historical data, E nm is the normalized anomaly score.

[0117] This formula is used to normalize the anomaly score E to the interval [0,1], E min and E max are the minimum and maximum values ​​of the anomaly score in the historical data, and the normalized anomaly score E nm This makes the score unaffected by the original score range, making it easier to compare and process. nm Similarly, the closer the value is to 1, the higher the abnormality;

[0118] It should also be noted that after completing the matching of the new feature vector with the fault voiceprint database, once an abnormal voiceprint pattern is identified (i.e., S(i) < T8), the system should not only immediately trigger the alarm mechanism to notify the operation and maintenance personnel to intervene in the inspection, but also automatically start a more detailed fault diagnosis process, and combine other monitoring data (such as vibration, temperature, etc.) for multi-dimensional cross-validation to further confirm the fault type and severity. At the same time, the system should record the newly identified abnormal pattern characteristics as the future fault pattern library F1, F2, …, F N potential updates to ensure the timeliness and comprehensiveness of the database.

[0119] In addition, for the normalized anomaly score E nm Although the meaning of the score in the interval [0,1] is clarified, it does not mention how to use this score to guide the formulation of maintenance strategies. In fact, according to E nm The level of maintenance response plans can be set in advance. nm When it is close to 1, it indicates a serious abnormality and requires emergency shutdown and maintenance; nm If the risk is low, online monitoring or planned maintenance measures may be taken to achieve hierarchical management of fault response, optimize resource allocation, reduce unnecessary downtime, and ensure the stable operation and economic benefits of the wind farm.

[0120] Dot product of vectors , which is used to calculate the cosine of the angle between two vectors, thereby obtaining the similarity between them.

[0121] S5. Perform preliminary fault location based on abnormal soundprint patterns and dynamic adaptive wind energy models.

[0122] Furthermore, based on the anomaly score Enm And wind energy environmental quality coefficient Q, to obtain the comprehensive evaluation model, the expression is:

[0123] ;

[0124] Among them, L is the comprehensive evaluation score, w1 is the abnormal score E nm , w2 is the weight factor of the complement (1-Q) of the wind energy environmental quality factor Q;

[0125] Three thresholds are defined: safety threshold L1, warning threshold L2, and danger threshold L3. Based on the comprehensive evaluation score L, the fault is located into predefined categories through threshold division. If L<L1, it is classified as normal; if L1≤L<L2, it is classified as attention; if L2≤L<L3, it is classified as warning; if L≥L3, it is classified as danger.

[0126] It should also be noted that the setting of the safety threshold L1, warning threshold L2 and danger threshold L3 is based on a comprehensive consideration of historical data analysis, expert experience, equipment specifications, failure consequence assessment and industry standards. These thresholds are designed to evaluate the health status of the equipment in real time by monitoring the comprehensive evaluation score L, where L1 represents the safety threshold of the equipment within the normal operating range, L2 indicates that attention should be paid to the initial abnormality of the equipment, and L3 is the warning limit for the equipment in a high-risk state. By dynamically monitoring L and comparing these preset thresholds, the system can issue an alarm in time and guide maintenance personnel to take corresponding measures, thereby effectively preventing failures and ensuring the safe and efficient operation of wind turbines.

[0127] S6. Comprehensively analyze the preliminary fault location and expected performance indicators, confirm the fault location and type, and generate a fault report based on the impact of environmental changes on fault identification;

[0128] Furthermore, a fault diagnosis model based on multi-feature fusion is introduced to combine the comprehensive evaluation score L and the normalized anomaly score E nm , wind energy environment quality factor Q and specific index F of fault soundprint characteristics n sp Combined to form a multi-dimensional feature vector, the expression is:

[0129] ;

[0130] Among them, F n sp It is a specific indicator of the new feature vector. This part focuses on specific attributes in the newly collected data, such as the frequency distribution and energy distribution of the voiceprint features, which are key details for fault type identification;

[0131] Introducing a correction factor based on wind energy environmental quality factor Q , reducing the misjudgment rate of fault identification due to environmental factors, the expression is:

[0132] ;

[0133] in, is a scalar amplification factor, β is a parameter that determines the steepness of the curve, Q0 is a specific environmental quality standard, is the base of natural logarithms;

[0134] The correction factor Applied to the comprehensive evaluation score L, the corrected comprehensive score is calculated as follows:

[0135] ;

[0136] Among them, L' is the corrected comprehensive score;

[0137] The difference ΔL between the comprehensive evaluation score L and the corrected comprehensive evaluation score L' was calculated, and the linear relationship between ΔL and each environmental parameter was tested using the Pearson correlation coefficient;

[0138] For each environmental parameter, draw a scatter plot of ΔL and the corresponding environmental parameter, add a trend line on the scatter plot to intuitively show the trend of ΔL changing with the environmental parameter;

[0139] Based on the scatter plot, determine the environmental parameters that have the greatest impact on ΔL, and how changes in environmental parameters affect ΔL, and generate a fault report. It should also be noted that the corrected scoring result L' is used as the core indicator. The scoring combines equipment operation data and historical failure modes. By comparing L' with the preset failure threshold, the system automatically classifies the fault level;

[0140] In the process of analyzing the impact of changes in environmental parameters on the modified comprehensive score ΔL, the correlation coefficient and the slope of the trend line provide key insights. Specifically, the positive correlation indicates that whenever the environmental parameter value increases, the modified score ΔL tends to increase, which means that the deterioration of environmental conditions may increase the fault identification sensitivity or false alarm rate of the equipment. On the contrary, the negative correlation indicates that the increase in environmental parameters leads to a decrease in ΔL, which may mean that in some environments, the fault identification ability of the equipment is enhanced, or that environmental factors play a certain compensatory role and reduce the possibility of false alarms. By observing the slope of the trend line, we can quantify the strength and direction of this relationship. The positive slope confirms the positive connection between environmental parameters and ΔL, while the negative slope reveals the inverse relationship, providing data support for the optimization of maintenance strategies and fault prediction models of wind turbines.

[0141] The present embodiment also provides an intelligent fault identification system for a wind turbine generator set, comprising: a data acquisition module, a voiceprint data processing module, a fault voiceprint database management module, a fault identification and positioning module, and a fault report generation module; the data acquisition module is used to collect acoustic and environmental data of the wind turbine in real time, and transmit it to the control center via a wireless network; the voiceprint data processing module is used to denoise and extract features from the collected voiceprint data, and prepare pure feature information for fault diagnosis; the fault voiceprint database management module is used to maintain a comprehensive fault voiceprint feature library to support the learning and storage of new fault modes; the fault identification and positioning module is used to compare the processed voiceprint features with the database, and combine model prediction to quickly identify and locate potential faults; the fault report generation module is used to integrate analysis results, automatically generate a report containing fault details, location and recommended countermeasures, and assist in decision-making.

[0142] This embodiment also provides a computer device, which is suitable for the case of an intelligent fault identification method for a wind turbine generator set, comprising: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the intelligent fault identification method for a wind turbine generator set as proposed in the above embodiment.

[0143] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0144] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for realizing intelligent fault identification of a wind turbine generator set as proposed in the above embodiment is implemented; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, referred to as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, referred to as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, referred to as EPROM), programmable read-only memory (Programmable Red-Only Memory, referred to as PROM), read-only memory (Read-Only Memory, referred to as ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0145] In summary, the present invention achieves intelligent fault identification and environmental adaptability optimization of wind power equipment by installing high-sensitivity sensors and environmental monitoring equipment at key locations of wind turbines, combined with advanced voiceprint processing, machine learning and deep learning technologies. From real-time data collection to intelligent analysis of voiceprint features, to dynamic prediction of wind power performance and precise fault location, the present invention greatly improves the accuracy and timeliness of fault warnings, while ensuring the robustness of diagnostic results with the help of environmental correction factors. Through this series of comprehensive strategies, not only the operational efficiency of wind farms and the scientific nature of maintenance decisions are enhanced, but also the intelligent response to complex environmental conditions is promoted, providing strong technical support for the sustainable development of the wind energy industry.

[0146] Example 2, referring to Table 1, is the second example of the present invention. To further verify the advancement of the present invention, experimental simulation data of the intelligent fault identification method for a wind turbine generator set are provided.

[0147] Four wind turbines (A, B, C, and D) were selected as test objects, among which unit A applied traditional fault identification technology as the control group, while units B, C, and D respectively adopted technical improvements of different stages of the present invention, and the parameters included voiceprint data signal-to-noise ratio (SNR), environmental parameter accuracy (EPA), fault identification rate (FIR), response time (RT), and environmental adaptability index (EAI).

[0148] First, highly sensitive acoustic sensors and environmental monitoring equipment (including temperature, humidity, and wind speed sensors) were installed at key locations on the wind turbine generator set, and all data was synchronized in real time through a central data acquisition system.

[0149] Secondly, voiceprint noise reduction processing based on time domain filtering and FFT was implemented on unit B, and voiceprint features were extracted and classified through PCA and improved SVM model to build a fault voiceprint database.

[0150] Then, on this basis, unit C further integrates deep learning and long short-term memory network LSTM, adaptively adjusts model parameters according to real-time environmental parameters, and calculates wind energy environmental quality coefficient (Q) and expected performance index ( ).

[0151] Finally, unit D fully applies the present invention, combines voiceprint features with environmental adaptability models to identify and locate faults, and generates a fault report through a diagnostic model based on multi-feature fusion.

[0152] The details are shown in the following table:

[0153] Table 1 Experimental record table

[0154]

[0155] Through the analysis of the data in the above table, it can be clearly seen that the present invention significantly improves the fault identification capability of wind turbines, especially the noise reduction, response acceleration, and enhanced environmental adaptability achieved on units B, C, and D, demonstrating the advantages of fast and accurate fault location. For example, unit D has a response time of 30 seconds and a fault identification rate of 98%, highlighting the beneficial effects of improving operation and maintenance efficiency and ensuring a stable supply of wind power, providing strong technical support for the intelligent transformation of the wind energy industry.

[0156] The wind turbine intelligent fault identification method of the present invention not only realizes technical innovation, but also demonstrates its significant advantages in practice in improving the efficiency of wind turbine fault identification, enhancing environmental adaptability and optimizing maintenance strategies, thus providing a powerful solution for the intelligent operation and maintenance of the wind energy industry.

[0157] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the present invention.

Claims

1. A method for intelligent fault identification of wind turbine generator sets, It is characterized in that include: Wind turbines are equipped with acoustic sensors and environmental monitoring equipment to collect soundprint data and environmental parameters in real time; Perform noise reduction and feature extraction on voiceprint data to obtain voiceprint features; Use voiceprint features to establish a fault voiceprint database; Based on deep learning, a dynamic adaptive wind energy model and real-time collected environmental parameters are constructed, the parameters of the dynamic adaptive wind energy model are adaptively adjusted, and the wind energy environmental quality coefficient and the expected performance indicators of the wind turbine generator set under the current environment are calculated; Match the extracted voiceprint features with the fault voiceprint database to identify abnormal voiceprint patterns; Perform preliminary fault location based on abnormal soundprint patterns and dynamic adaptive wind energy models; Comprehensively analyze the preliminary fault location and expected performance indicators, confirm the fault location and type, and analyze the impact of environmental changes on fault identification; The specific steps of constructing a dynamic adaptive wind energy model based on deep learning and real-time collected environmental data, adaptively adjusting the parameters of the dynamic adaptive wind energy model, and calculating the wind energy environmental quality coefficient and the expected performance index of the wind turbine generator set under the current environment are as follows: The dynamic adaptive wind energy model is constructed by combining deep learning and long short-term memory network LSTM to process time series data. The expression is: ; in, represents the expected performance indicator at time t, represents the parameter set of the dynamic adaptive wind energy model, V(t) is the wind speed time series, T(t) is the temperature time series, H(t) is the humidity time series, and θ(t) is the wind direction time series; According to the learning mechanism in the model, the parameters are constantly fine-tuned according to new real-time data. The expression is: ; in, is the learning rate, is the gradient of the loss function with respect to the parameters of the long short-term memory network LSTM, is the actual measured power output, are the updated model parameters, are the model parameters before updating, represents the expected performance indicator at time t; Based on model training and adaptive adjustment, the wind energy environment quality coefficient Q is defined as a composite indicator, and the expression is: ; Where V is wind speed, θ is wind direction, T is temperature, H is humidity, is a function of wind speed and wind direction, β is the temperature-related attenuation coefficient, γ is the humidity-related attenuation coefficient, T 0 is the reference temperature, η is a constant used for scaling, is the base of natural logarithms; Using the trained dynamic adaptive wind energy model, the current real-time environmental parameters are input, and the model output is the expected performance index of the wind turbine under the current environment. ,The output calculation of the model relies on the forward propagation mechanism of the long short-term memory network LSTM, which maps the input environmental features to the predicted values ​​of the expected performance through a series of nonlinear transformations and weight adjustments.

2. The intelligent fault identification method for a wind turbine generator set according to claim 1, Features: The wind turbine generator set is equipped with acoustic sensors and environmental monitoring equipment to collect voiceprint data and environmental parameters in real time. The specific steps are as follows: Voiceprint data and environmental parameters are collected in real time through high-sensitivity acoustic sensors with anti-interference capabilities installed at key locations of wind turbines and environmental monitoring equipment installed around wind turbines. All equipment is connected to the central data acquisition system.

3. The intelligent fault identification method for wind turbine generator set according to claim 2, Features: The voiceprint data is subjected to noise reduction and feature extraction to obtain voiceprint features; and a fault voiceprint database is established using the voiceprint features. The specific steps are as follows: The time domain filter is used to directly process the time series data of the voiceprint data, and then combined with the frequency domain analysis, FFT is used to convert the voiceprint data into a frequency domain signal. Spectral subtraction is used in the frequency domain to remove noise of specific frequencies. Finally, IFFT is used to convert the processed frequency domain signal back to the time domain to obtain the denoised voiceprint data. The expression is: ; Where X[k] is the kth frequency component in the frequency domain, x[n] is the input voiceprint data, k is the index of the frequency component, n represents the nth sample, and FFT means accepting a real number as input and returning the corresponding frequency domain representation; The voiceprint features are extracted by calculating the RMS value and peak value in the time domain, spectrum analysis in the frequency domain, and Mel frequency cepstral coefficients; A multidimensional feature vector is constructed based on the extracted voiceprint feature vector, and principal component analysis (PCA) is used to reduce the feature dimension, eliminate redundant information and highlight key features. The expression is: ; Among them, X is the original voiceprint feature vector, P is the conversion matrix, Y is the voiceprint feature vector after dimensionality reduction, and T is the matrix transpose operator; A support vector machine (SVM) model is trained using the voiceprint feature vector after dimensionality reduction, and an adaptive regularization parameter is introduced to improve the SVM model, allowing the model to automatically adjust the regularization strength according to the misclassification cost of different fault types. The expression is: ; in, is the weight vector, b is the bias term, Ci is the adaptive regularization parameter, is a slack variable, m is the number of samples, and i is an index variable; The improved SVM model is used to classify the real-time monitoring data into faults. According to the classification results, the corresponding voiceprint features and their fault types are stored in the fault voiceprint database. The established fault voiceprint database is used to perform real-time analysis on the newly collected voiceprint data. By comparing the voiceprint features in the database, early fault detection and early warning are achieved. Trend analysis and fault cause backtracking are performed based on historical fault data. The expression is: ; Among them, F t+1 is the fault prediction value at time t+1, is the smoothing coefficient, is the failure prediction value at time t, is the fault prediction value at time t-1.

4. The intelligent fault identification method for wind turbine generator set according to claim 3, Features: The extracted voiceprint features are matched with the fault voiceprint database to identify abnormal voiceprint patterns. The specific steps are as follows: Assume that the fault mode feature set of the fault voiceprint database is {F1, F2, …, FN}, and each feature vector F in the fault mode feature set is i Represents a known fault mode, and uses cosine similarity as the matching metric to calculate the similarity between the new feature vector and each vector in the database. The expression is: ; Among them, S(i) represents the similarity between the new feature vector and the i-th fault mode feature vector, is the dot product of the vectors, F new represents the new eigenvector, F i is the feature vector of the i-th known fault mode in the fault voiceprint database; Define an abnormal threshold T 8 , when S(i)< T 8 When , it is considered that the newly collected voiceprint data does not match any known fault mode, and a new abnormal mode appears. A similarity-based abnormal scoring function is introduced, and the expression is: ; Among them, E is the preliminary abnormality score, (S(i)) represents the new feature vector F new and all the fault mode feature vectors F in the database i The maximum value of the cosine similarity; The initial anomaly score is normalized and the expression is: ; Among them, E min is the minimum value of the anomaly score in the historical data, E max is the maximum value of the anomaly score in the historical data, E nm is the normalized anomaly score.

5. The intelligent fault identification method for wind turbine generator set according to claim 4, Features: The specific steps of performing preliminary fault location based on abnormal soundprint patterns and dynamic adaptive wind energy models are as follows: Based on the anomaly score E nm And wind energy environmental quality coefficient Q, to obtain the comprehensive evaluation model, the expression is: ; Among them, L is the comprehensive evaluation score, w 1 is the anomaly score E nm The weight factor, w 2 It is the weighting factor of the complement (1-Q) of the wind energy environmental quality factor Q; Define three thresholds: safety threshold L 1 , the warning threshold is L 2 , the danger threshold L 3 , based on the comprehensive evaluation score L, the fault is located in a predefined category through threshold division. If L<L 1 , it is classified as normal. If L 1 ≤L<L 2 , then it is classified as attention, if L 2 ≤L<L 3 , it is classified as a warning. If L ≥ L 3 , it is classified as dangerous.

6. The intelligent fault identification method for wind turbine generator set according to claim 5, Features: The preliminary fault location and expected performance indicators are comprehensively analyzed to confirm the fault location and type, and analyze the impact of environmental changes on fault identification. The specific steps are as follows: A fault diagnosis model based on multi-feature fusion is introduced, which combines the comprehensive evaluation score L and the normalized anomaly score E nm , wind energy environment quality factor Q and specific index F of fault soundprint characteristics n sp Combined to form a multi-dimensional feature vector, the expression is: ; Among them, F n sp is the specific index of the new eigenvector; Introducing a correction factor based on wind energy environmental quality factor Q , reducing the misjudgment rate of fault identification due to environmental factors, the expression is: ; in, is a scalar amplification factor, β is a parameter that determines the steepness of the curve, and Q 0 is a specific environmental quality standard. is the base of natural logarithms; The correction factor Applied to the comprehensive evaluation score L, the corrected comprehensive score is calculated as follows: ; Among them, L' is the corrected comprehensive evaluation score; The difference ΔL between the comprehensive evaluation score L and the corrected comprehensive evaluation score L' was calculated, and the linear relationship between ΔL and each environmental parameter was tested using the Pearson correlation coefficient; For each environmental parameter, draw a scatter plot of ΔL and the corresponding environmental parameter, add a trend line on the scatter plot to intuitively show the trend of ΔL changing with the environmental parameter; Based on the scatter plot, determine the environmental parameters that have the greatest impact on ΔL, how changes in environmental parameters affect ΔL, and generate a fault report.

7. A wind turbine generator set intelligent fault identification system, based on the wind turbine generator set intelligent fault identification method according to any one of claims 1 to 6, Features: Including data acquisition module, voiceprint data processing module, fault voiceprint database management module, fault identification and positioning module, fault report generation module; The data acquisition module is used to collect acoustic and environmental data of the wind turbine in real time and transmit it to the control center via a wireless network; The voiceprint data processing module is used to perform denoising and feature extraction on the collected voiceprint data to prepare pure feature information for fault diagnosis; The fault voiceprint database management module is used to maintain a comprehensive fault voiceprint feature library to support the learning and storage of new fault modes; The fault identification and location module is used to compare the processed voiceprint features with the database, combined with model prediction, to quickly identify and locate potential faults; The fault report generation module is used to integrate the analysis results and automatically generate a report containing fault details, location and recommended countermeasures to assist decision-making.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program. Features: When the processor executes the computer program, the steps of the wind turbine generator set intelligent fault identification method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, Features: When the computer program is executed by a processor, the steps of the wind turbine generator set intelligent fault identification method according to any one of claims 1 to 6 are implemented.